6.4 Variability of values across different household groups. 45 ..... In the Great Southern, the impacts of land degradation are already obvious while in the Fitzroy ...
Estimating community values for land and water degradation impacts Final Report
Prepared for the National Land and Water Resources Audit, Project 6.1.4 Martin van Bueren and Jeff Bennett
November 2000
Contents Contents
1
Chapter 1: Introduction
3
Chapter 2: Analytical approach 2.1 The valuation task 2.2 Choice modelling Technique overview Technique strengths
5 5 6 6 7
Chapter 3: Research issues 3.1 Definition and selection of attributes Attribute definition Selection criteria 3.2 Communication aspects 3.3 Benefit transfer issues Framing effects Population effects Minimising transfer error 3.4 Framing and sampling strategy
9 9 9 10 10 11 11 12 12 13
Chapter 4: Questionnaire design and administration 4.1 Survey of scientists and resource managers 4.2 Focus groups Environmental awareness Attribute selection and definition Responsibilities and funding mechanism. 4.3 Development of choice options 4.4 Choice set design Marginal versus absolute format Option labels versus generic options Attribute icons 4.5 Pre-testing 4.6 Sampling 4.7 Survey administration
15 15 15 15 16 17 18 23 23 23 25 25 25 25
Chapter 5: Descriptive overview of survey results 5.1 Response rate Overall response State differences in response rate. 5.2 Completion of the choice task. 5.3 Description of data Sample characteristics Representativeness of the national sample 5.4 Preliminary assessment of willingness to pay 5.5 Model specification and parameter estimates Specification Parameter estimates
27 27 27 28 30 30 30 32 34 36 36 38
1
Chapter 6: Value estimates from the national survey 6.1 Overview 6.2 Attribute implicit prices 6.3: Welfare impacts from alternative scenarios Biodiversity protection scenario Waterway restoration scenario Negative social impacts scenario Positive social impacts scenario 6.4 Variability of values across different household groups 6.5 Aggregate welfare impacts of resource use change
41 41 41 43 43 44 44 44 45 47
Chapter 7: Transferability of value estimates 7.1 Overview 7.2 Benefit transfer tests BT Test 1: Transferability of estimates from a national to regional context BT Test 2: The relative importance of framing BT Test 3: Consistency of values across case study regions BT Test 4: Consistency of values across city and regional respondents 7.3 Conclusions
49 49 49 49 50 52 52 54
Chapter 8: Benefit transfer guidelines 8.1 Overview 8.2 Implicit price transfer 8.3 Choice model transfer
55 55 55 57
References Appendix A: Script for the focus group discussions Appendix B: Attribute implicit prices Appendix C: Background information accompanying the national survey
59 61 62 63
2
Chapter 1: Introduction The purpose of this study is to estimate dollar values for non-market environmental and social impacts that are associated with land and water degradation in Australia. It provides quantitative information about the size of trade-offs between different social and environmental outcomes that stem from different resource use decisions. The study emerges out of a need to understand how the Australian community values goods and services that are not exchanged in markets. A better knowledge of these values will assist resource managers to make more informed policies based on a comprehensive set of costs and benefits associated with resource use changes. The Project focuses on both “use values ” and “non-use values”. Examples of use values include outdoor recreation and the passive enjoyment of scenic beauty. Non-use values refer to benefits that society obtains from environmental resources in the absence of any tangible, current interaction with the resource. For instance, individuals may benefit from knowing that a natural area exists in an intact, “healthy” state even if they never intend to visit the area. Similarly, a non-use benefit may stem from the knowledge that country communities are in a viable and prosperous state. Together, these use and non-use values contribute to the total non-market impacts associated with a change in resource use. A key objective of the study is to produce value estimates for a set of generic attributes that characterise the environmental and social impacts of land and water degradation at national and regional levels. The goal is for these attribute value estimates to be transferable across different regions and populations within Australia (a practice known as benefit transfer). The concept of transferability is appealing because it overcomes the need to undertake expensive surveys each time a new project proposal is evaluated, and is consistent with the rapid assessment approach being promoted by the Audit. However, the practice can lead to significant errors if the source values obtained from a pre-existing study are context-dependent and that context does not match the conditions which prevail at the target area of interest (Brouwer, 2000). Thus, an important component of this study is an investigation of the conditions and limits that apply to benefit transfer, and the development of a systematic procedure for calibrating value estimates so that they can be validly transferred from one policy context to the next. A survey technique known as Choice Modelling is used in this study to estimate attribute values and welfare impacts for alternative resource use scenarios. It is the preferred valuation method because it is particularly suited to the role of providing value estimates that can be used as a source of data for benefit transfer. Relative to Contingent Valuation, it enables better control over the frame of reference within which non market goods are presented to respondents for valuing. It also enables the total value of a resource use change to be disassembled into its component attributes. The report is organised as follows. Chapter 2 contains an overview of the Choice Modelling technique and the challenges of estimating non-market values. Chapter 3 summarises the main research issues that underpin this study, namely the selection of appropriate attributes and the factors that complicate benefit transfer, including framing, scope, and population effects. Chapter 4 contains a detailed description of how the questionnaire was designed and administered. In Chapter 5, a descriptive review of the main results is given. This is followed by an in-depth examination of the national value estimates in Chapter 6. The results of a 3
number of benefit transfer tests are summarised in Chapter 7. These tests investigate the validity of transferring attribute value estimates from one policy context to another. The report concludes, in Chapter 8, with a set of guidelines and recommendations for using the value estimates to assess the non-market impacts of regional and national policies.
4
Chapter 2: Analytical approach 2.1 The valuation task In this study the concept of value is treated from an economic perspective. Economists define value in terms of the maximum amount an individual is willing to pay for a good or service less the price paid for that good or service. It is assumed in welfare economics that individuals consume goods with the objective of maximising their wellbeing (or utility), subject to a budget constraint. This assumption holds for both marketed and non-market goods. The strength of people’s concerns for the environment, and ethics, are encapsulated by this definition of wellbeing. If the theory of utility maximisation is embraced, it is possible to express all values in terms of a standard money-metric, namely an individual’s “willingness to pay”. The task of estimating values for environmental and social impacts is challenging because many of these “goods” are not exchanged through markets. Consequently, market price and demand information is not available. Instead, non-market valuation techniques must be used to estimate the preferences and values of individuals. A variety of non-market valuation methods have been developed for estimating the amount an individual is willing to pay for improvements in environmental or social outcomes. These methods produce marginal values because they concentrate on the value of incremental changes in the level of an outcome. There are two categories of non-market valuation techniques: Revealed preference and stated preference methods. The former uses observations of people’s behaviour to infer values for environmental goods. Examples include visits to recreation sites (the travel cost method) or the selection of residential locations in close proximity to scenic views (the hedonic price technique). Revealed preference techniques are useful for estimating use values but are not capable of estimating non-use values. As non-use values are an important component of this study, a stated preference method was adopted. Stated preference techniques involve asking respondents about their maximum willingness to pay for a specific change in the supply of a non-market good. The Contingent Valuation Method is one such technique. It has been used in a number of prominent Australian studies for valuing environmental resources. Perhaps the best known of these is a study undertaken by the Resource Assessment Commission to assess the environmental costs of mining at Coronation Hill near Kakadu National Park (Imber, Stevenson and Wilks, 1991). Other studies of national significance include an estimation of forest conservation benefits on Fraser Island (Hundloe et. al. 1990) and an assessment of soil erosion costs in New South Wales (Sinden, 1987). This study employs an alternative stated preference technique known as Choice Modelling. The technique originates from the marketing and transport literature where it has been used extensively to analyse consumers' choices of products and transport modes, respectively. It has only recently begun to be used by economists for valuing environmental impacts.
5
2.2 Choice modelling Technique overview
In a Choice Modelling (CM) application, respondents are presented with a series of questions, each containing a set of options known as a choice set. Typically, five to eight choice sets are included in a questionnaire. In each choice set, respondents are asked to choose their preferred option from a range of alternatives. Figure 2.1 contains an example of a choice set that was used in a study of wetland rehabilitation (Morrison, Bennett and Blamey 1999). The options can be viewed as outcomes from alternative management policies, which are described in terms of a standard set of attributes or characteristics. Just as a car has a number of distinct attributes that contribute to its appeal (eg. air-conditioning, colour, fuel economy, price), each resource use option in an environmental valuation choice set is described by a number of key attributes and their associated levels. In a CM application, the options making up the choice sets are formed by allowing attribute levels to vary systematically according to an experimental design. Each choice set also includes a status quo option that describes the outcome that is associated with a “no change” policy. It serves as a base against which to measure respondents’ willingness to make tradeoffs in securing change. The other options are deviations from the status quo. Figure 2.1: An example of a choice set and its key components (from Morrison, Bennett and Blamey). Attribute levels Attributes
Your water rates
Option A
Option B
Status Quo
$150 increase
$20 increase
No change
Wetlands area
800 km
550 km
400 km2
Waterbirds breeding frequency
every 3 years
every 2 years
every 6 years
Number of native fish species
25 species
12 species
5 species
Irrigation-related employment
2000 jobs
1500 jobs
2800 jobs
I would choose:
2
Option A Option B
2
Respondent’s choice of option
Status Quo
The data collected from people’s responses to the choice questions reveal the extent to which individuals are prepared to trade-off one attribute against another (see Box 2.1 for detail on the theory that underpins Choice Modelling). Provided one of the attributes is measured in dollar terms, it is possible to estimate the amount of money people are prepared to pay for improving a non-monetary attribute by one unit. This value is known as an implicit price. The money attribute used in the choice sets can take the form of a tax, licence fee, entry fee, or some other payment mechanism.
6
In addition to implicit prices, the CM technique enables welfare impacts to be calculated for various resource use scenarios. Valuation is not restricted to the set of scenarios presented in the questionnaire. Rather, the costs or benefits associated with a whole range of change scenarios can be calculated once parameters have been estimated for the choice model. The CM application need only employ a range of attribute levels sufficient to cover the range of scenarios that are of interest. The technique can also be used to examine the level of non-monetary community support for alternative policies that have specific outcomes. Support is measured in terms of the proportion of respondents who would choose a particular policy. This type of information can be useful for gauging the relative popularity of various strategies among different stake-holder groups.
Box 2.1: Underpinning theory of Choice Modelling The choice behaviour of respondents is assumed to be underpinned by a theory known as Random Utility Theory. The utility obtained by individual i from choosing alternative j in a choice set is given by:
Vij = ( q j , c j , si ,ε ij ) where qj is a vector of quality attributes, cj is the cost of the alternative (given by the levy attribute), sj is a vector of the individual’s socioeconomic characteristics, and εij is an error term. An error term is included to reflect the fact that the researcher does not know all the factors that contribute to an individual’s utility. The probability of individual i choosing alternative j is given by:
Prij = Pr[{vij ( q j , c j , si ) + ε ij } ≥ {vik ( qk , ck , si ) + ε ik }] ∀ j ≠ k This equation says that the probability of a respondent choosing alternative j is equal to the probability that the utility associated with that alternative exceeds the utility associated with any other alternative k in the choice set. The random utility model is made operational by adopting a particular cumulative density function for the unobserved component of utility, ε. If the ε's are independently and identically distributed with a extreme value type I (Weibull) distribution, then the individual's probability of choosing site j is given by a multinomial logit model (McFadden 1974):
Pr j =
exp(v j ) J
∑ exp(v ) k =1
k
Parameters of the utility function are estimated by Maximum Likelihood, which finds values for the coefficients that maximise the likelihood of the pattern of choices in the sample of observations. In this study, the software package LIMDEP (Greene, 1995) was used to estimate the multinomial logit model.
Technique strengths
Choice Modelling was selected as the preferred method for this analysis because it has a number of potential strengths over Contingent Valuation:
7
•
It forces respondents to consider trade-offs between attributes.
•
It makes the policy frame explicit to respondents via the inclusion of an array of options.
•
It allows the estimation of implicit prices for attributes.
•
It has the flexibility of being able to estimate welfare impacts for multiple scenarios.
•
It has the capability to estimate the level of community support for alternative scenarios in non-monetary terms.
•
It enables the total value estimate of a resource use change to be disassembled into its component parts (attributes), which facilitates benefit transfer.
•
It potentially reduces the incentive for strategic behaviour.
8
Chapter 3: Research issues This section of the report contains a brief overview of the research issues that are tackled by this study. Section 3.1 discusses the alternative ways that attributes could be defined and the criteria that were used to select the attributes. In Section 3.2, communication aspects of the CM questionnaire are considered. It includes a discussion of the steps taken in this study to reduce the cognitive burden placed on respondents. Section 3.3 deals with issues relating to benefit transfer, namely sources of transfer error, frame manipulation and the aggregation of benefit estimates. The chapter concludes with an outline of the approach used in this study to investigate the effect of framing and population characteristics on value estimates. 3.1 Definition and selection of attributes The selection of an appropriate set of attributes that best reflect the impacts of land and water degradation is a critical component of this Project. It entails categorising the physical outcomes of any given resource use scenario into separate components. This task is not straight-forward because environmental impacts are inherently complex and interrelated. The attributes need to be sufficiently generic so that they are capable of describing a wide variety of resource-use outcomes at different regions of Australia. They also need to be relevant to the public whilst being measurable and objective. The task of defining attributes is complicated by the added requirement that they be independent and not causally related. Attribute definition
At least two alternative approaches can be taken to defining the attributes. One possibility is to describe environmental impacts in terms of "degradation issues" (eg. salinity, soil erosion, pests). Using this approach, the area of salinity would be regarded as an environmental attribute. It tends to be consistent with the way resource managers compartmentalise policy outcomes and set priorities for future work. An alternative method would be to move from an environmental issues focus to one that is based on biophysical impacts. This requires a concentration on specific biophysical factors such as changes in species diversity and fish abundance. Defining attributes in terms of biophysical impacts offers a number of advantages. Firstly, people are usually more concerned about the way in which degradation might affect the things they cherish rather than the processes causing the changes. For the purposes of CM, it is important to define the attributes in terms that are meaningful to respondents. Secondly, biophysical impacts tend to be more generic than environmental issues and degradation processes. This is because different forms of degradation often share common biophysical impacts (Figure 3.1). Consequently, it is possible to apply one standard set of attributes to describe the impacts of multiple forms of degradation, irrespective of geographic location. For instance, the impact of degradation on endangered species can be expressed generically, regardless of whether losses are caused by dryland salinity or remnant vegetation clearance.
9
Figure 3.1: A diagram showing the distinction between degradation issues and biophysical impacts.
Weeds & pests
Biophysical impacts
Biophysical impacts common to all three forms of degradation
Biophysical impacts
Dryland salinity
Biophysical impacts
Soil erosion
Selection criteria
A number of criteria were used in this study for selecting attributes. One of the primary requirements of CM is that respondents must perceive attributes to be independent of one another. Meeting this condition is difficult because many environmental impacts are interrelated. For example, an attribute defined as “area of healthy remnant vegetation” could be causally prior to “species diversity”, meaning that healthy remnant vegetation may be viewed as a necessary prerequisite for supporting species diversity. Respondents may value both attributes but there is the possibility that less weight will be given to species diversity if it is believed that native vegetation must be restored first. Causal attributes complicate the modelling of choice behaviour. Previous research has shown that when a causally prior attribute is included in a questionnaire, the value estimated for the “downstream” attribute is depressed relative to the estimate obtained when the causal attribute is omitted (Blamey, Bennett, Morrison, Louviere and Rolfe1998). Therefore, causality should be minimised by omitting either the causal attribute or the downstream attribute. The choice of which one to omit depends largely on how the value estimates are to be used. Other criteria for selecting attributes include the need to ensure that attributes are meaningful to respondents, quantifiable, and of relevance to decision-makers. It is critical that attributes have common interpretation among all respondents. Poorly defined attributes may prompt some respondents to value a wider array of goods than those intended by the researcher. 3.2 Communication aspects Compared to Contingent Valuation, a CM questionnaire is longer and more complex. It requires respondents to process a large amount of information including: •
Background information relating to the issues and scenarios.
•
A series of five or more choice decisions that involve multiple options. 10
•
Different combinations of attribute outcomes under each option.
•
Combinations of attribute outcomes that may appear counter intuitive to respondents.
•
A large array of numerical information associated with the options.
Owing to the considerable cognitive burden this process places on respondents, it is important to design the CM questionnaire so that it communicates the choice task to respondents as clearly as possible. Previous CM research has found that respondents use various ways of simplifying the choice task. For instance, "heuristics" may be employed whereby choices are made on the basis of one or two "indicator" attributes with no attention paid to the other attributes. This behaviour is clearly undesirable because the intention of CM is to encourage respondents to weigh up the options based on an appraisal of all the attributes. In an effort to improve respondent cognition, this study adopted visual stimuli as a means of denoting the attributes and their levels. These graphics were intended to reduce the complexity of the choice task and improve the communication of attribute outcomes. 3.3 Benefit transfer issues An important goal of this study is to estimate values for a set of attributes that can then be transferred to a “target” region and used to evaluate the non-market costs and benefits of public sector investment in different projects and policies. Whilst the concept of transferring “off the shelf” estimates to particular regions of interest is appealing, the validity of this practice is restricted to cases where there is a reasonable degree of similarity between the source study and target area. Framing and population differences could render the estimates from a source study to be inappropriate for informing policy at a target site. Framing effects
The term frame is used to describe the way in which aspects of a situation influence people’s involvement in, and experience of, the situation. Therefore, when an individual is asked about his or her willingness to pay for a particular environmental improvement, the environmental “good” is embedded in a frame. Some important elements of the frame include: •
the scope of changes in resource use under investigation;
•
the array of substitute and complementary goods;
•
the institutional setting; and
•
questionnaire cues.
In order to transfer benefit estimates from one context to another it is necessary to gain and insight into how different frames influence people’s values. Embedding is one aspect of framing. Embedding effects are said to occur when respondents are willing to pay more for a good when it is assessed individually compared to when it is valued as part of a more inclusive package. For example, respondents may be willing to pay $150 to protect 50 hectares of remnant vegetation when offered as a single outcome, and only $15 for that same area of bush when offered as part of a bundle of environmental outcomes. This result is common in the non-market valuation literature. 11
The embedding effect is not an aberration or bias. Its presence is consistent with standard economic theory in that the value of a good is dependent on the range of substitute and complementary goods available to a consumer. Hence, the wider the array of substitutes, the lower the value of an individual good, while commodities that serve as complements generally enhance the value of a good. Thus, the frame in which a good is embedded is important for valuation. The challenge for the researcher is to ensure that the questionnaire frame is appropriate for the policy being investigated. Population effects
Population differences are another factor that could cause differences between values estimated in different regions. Values are likely to be sensitive to a population’s socioeconomic characteristics, attitudes and social norms. The cultural traditions of a region, and its institutions could also be important. The issue of population effects is explored in this study by comparing the results derived from the same estimation procedure being applied across a number of different population samples. Furthermore, values are estimated for a range of different household groups, categorised according to specific characteristics. Information from these analyses is incorporated into the benefit transfer guidelines. The eventual size of population to which value estimates are to be aggregated (known as the geographic extent of the market) is another important consideration when transferring benefits. The practice of aggregation is, itself, a form of benefit transfer if the source study estimates are derived from a sample that is different to the target population. Previous work has shown that non-market values, in particular use-values, frequently decline as the distance between a respondent’s residence and study site increases (Pate and Loomis, 1997; Sutherland and Walsh, 1985). The same relationship is less likely to hold for non-use values. A choice modelling study by Rolfe and Bennett (2000) found evidence of significant population differences within the state of Queensland. Using a split-sample test, it was shown that respondents who resided in rural areas have lower values for conserving remnant vegetation in the Desert Uplands of Central Queensland than metropolitan Brisbane residents. The implication of this finding is that values are not necessarily inversely proportional to distance. Community attitudes are also influential in determining values. Minimising transfer error
Differences in population characteristics and attitudes between the source and target regions can partly be accounted for by transferring a “value function” rather than point estimates of value. An example of a point estimate is the average "per person" value of an additional hectare of remnant vegetation. An alternative approach is to specify a value function which specifies an individual’s willingness to pay as a function of a number of explanatory variables including population characteristics. This procedure adjusts for some of the differences between source and target sites and has been shown to out-perform point transfers in tests of benefit transfer (Brouwer, 2000). However, the value function does not necessarily control for all the factors related to frame and population. Thus it was important in this study to quantify the sensitivity of value estimates to different frames and populations. This information provides a guide for calibrating source estimates so that they can be validly transferred to a different site or policy context. The approach taken was to design a framing and sampling strategy that allows the influence of 12
population and frame to be tested. 3.4 Framing and sampling strategy As the primary purpose of this research project is to develop a set of value estimates for later use in benefit transfer, it is important to gain a better understanding of the way in which frame, scope, and population differences interact to influence value estimates. A research strategy was developed to investigate the following questions: •
To what extent are community preferences and values dependent on the frame?
•
Are respondents sensitive to the scope of environmental impacts proposed in a questionnaire?
•
Do parochial attitudes play a significant role in influencing values?
•
How do community preferences and values change with distance from a study site?
•
What adjustments are needed if attribute value estimates are to be validly transferred from a national context to a regional context?
Specifically, the strategy involved the development of three separate questionnaire versions, each representing a different frame. One of the questionnaires focused on land and water degradation in a national context, whilst the other two dealt with degradation issues in two case study regions. The regions selected for the case studies were the Great Southern Region (GSR) of Western Australia and the Fitzroy Basin Region (FBR) of Central Queensland. The degradation issues in these regions are markedly different and there is evidence to suggest that Queensland people have different attitudes towards the environment to Western Australians1. Thus, the two regions were selected as a means of testing the transferability of the national estimates over a wide range of circumstances. The other component of the research design was the sampling strategy. The national questionnaire was issued to a random sample of the Australian population, while the case-study questionnaires were administered to households living in the vicinity of each region; one from the region’s main population centre and the other from the region’s state capital city population. The main population centres for the GSR and FBR are Albany and Rockhampton respectively. The corresponding capital cities are Perth and Brisbane. As depicted in Table 3.1, this framing and sampling strategy allows an investigation of seven different combinations of frame and population, resulting in seven separate choice models. A common set of attributes was used for all versions of the questionnaire and the same three levy amounts were used across all versions. However, the frame for each version was manipulated by adjusting the levels of the social and environmental attributes so as to match the conditions that exist in each case study area. In addition, the frame of reference was varied across the three different versions by tailoring the background information that accompanied
1
A survey by the Australian Bureau of Statistics (ABS) indicated that WA residents have a greater awareness of environmental problems than any of the other States, and Queenslanders have the lowest levels of awareness (ABS, Catalogue 4602, 1999). 13
the questionnaires. Respondents were provided with information that reflected the issues and policies that are relevant to each study area (see Appendix C for a copy of the information booklet that was sent out with the national survey). Table 3.1: Summary of the models estimated for various combinations of population and questionnaire frame. POPULATION Regional sample Rockhampton
FRAME
Fitzroy Basin
Model 5
Great Southern National
Albany
Brisbane
Perth
National sample National
Model 7 Model 4
Model 3
Capital city sample
Model 2
Model 6 Model 1
14
Chapter 4: Questionnaire design and administration 4.1 Survey of scientists and resource managers An initial list of environmental attributes was compiled by surveying approximately 35 scientists and resource managers. The purpose of this preliminary survey was to obtain a wide-ranging review of attributes that were considered to be important from the perspective of policy makers and their advisers. The questionnaire was framed at the national level. No reference was made to specific case study regions. This initial scoping survey indicated that resource managers find it difficult to differentiate between issues and biophysical outcomes. Nevertheless, the survey provided a starting point for identifying possible attributes. 4.2 Focus groups The next phase of questionnaire design involved structured focus group discussions with members of the public. In total, approximately 65 people attended seven focus group meetings over a period of two months. The meetings were held in the following locations: City
Regional
• • • •
• • •
Sydney. Canberra. Perth. Brisbane.
Yass, NSW. Rockhampton, Qld. Albany, WA.
The duration of each meeting was one and a half hours. Market research companies were contracted to recruit ten participants for each group. People from a cross section of the community were selected for the groups, ensuring a mix of genders, age groups (18-65 years), and occupational backgrounds. To prevent the groups from containing a disproportionate number of participants with a pro-environment disposition, care was taken not to divulge the topic of the discussions at the time of recruitment. Recruits were told that they would be helping to develop a questionnaire concerning social issues of national importance. The initial meetings were primarily used to gain an understanding of public awareness of environmental issues and to generate a list of environmental attributes. Of particular interest was whether people “think” at a local level or at a more general, national level. The meetings held in regional areas provided an insight into the aspirations of country people, and how these contrasted to the preferences of city dwellers. Another goal of the focus group work was to check communication aspects in early versions of the questionnaire. Appendix A contains a copy of the discussion questions that were used in these focus groups. Environmental awareness
The focus group work revealed that environmental issues are not given a high priority by rural or metropolitan communities relative to other social issues. This finding is consistent with an Australian Bureau of Statistics survey of households in which only nine per cent of households ranked environmental concerns as their top social issue (ABS, Catalogue 4602, 1999). People from the city focus groups generally had less knowledge of land and water degradation issues than people in the regional centres. They were aware of high-profile issues, such as salinity, through media coverage but they had little understanding of the causes and impacts of 15
degradation. Their greatest concern was the impact that degradation might have on human health via effects on water and food quality. A second-ranked concern was the possibility that degradation may increase the cost of food and water. Mention was also made of the need to maintain “viable country communities”. Fewer references were made to the impact of degradation on conservation values. Attribute selection and definition
The focus group discussions identified a set of concerns that were consistent across most focus groups, albeit with differing degrees of emphasis depending on the particular case study region. The five main categories of environmental and social concerns were: •
Native species and ecosystem functioning.
•
Landscape aesthetics.
•
Outdoor recreation opportunities.
•
Productivity of the land and quality of drinking water.
•
Viability of country communities.
Notably, these concerns comprise both use and non-use dimensions. The desire to preserve native species and to maintain viable country communities constitute non-use values. The demand for attractive landscapes, outdoor recreation areas, and the maintenance of production activities reflect use values. The list of concerns provided by the focus groups was used to define four attributes for the CM application, three of which were environmental and the fourth that captured peoples’ social concerns (Table 4.1). Production-related concerns were omitted from the choice model because a separate study within Theme 6 of the Audit estimates the cost of damage to agricultural production. Instead, respondents were asked to concentrate on the conservationrelated effects of degradation. Table 4.1: Environmental attributes selected for the choice modelling questionnaire. Attribute
Unit of measurement
Species Protection
The number of species protected from extinction.
Landscape Aesthetics
The area of farmland repaired and bush protected.
Waterway Health
The length of waterways restored for fishing or swimming.
Social Impact
The net loss of people from country towns each year.
Species Protection The Species Protection attribute was included to capture respondents’ non-use values for ecological protection. It was measured in terms of the number of endangered species protected from extinction under a particular resource use scenario. Landscape Aesthetics 16
Landscape aesthetics was measured in terms of “hectares of farmland repaired or bush protected”. This unit of measurement accommodates the differing circumstances of the two case study regions. In the Great Southern, the impacts of land degradation are already obvious while in the Fitzroy degradation remains largely a potential of further development. However, there are some possible drawbacks with defining landscape aesthetics in this way. The focus group discussions revealed that some people view the repair of farmland as a productionrelated activity and bushland protection as conservation-related. Others believed that the better management of the landscape to improve aesthetics would also protect endangered species (a problem of causality). In order to minimise the potential for causality and the risk of respondents broadening their valuations to production-related aspects land management, the aesthetics attribute was repeatedly referred to in the questionnaire as a measure of countryside attractiveness. Waterway Health The impact of degradation on recreational opportunities was defined by the Waterway Health attribute. This attribute was designed to capture respondents’ joint concerns for recreational activities and the preservation of waterway habitats. It was defined in terms of fishing and swimming opportunities so as to deflect attention away from the production values associated with water resources. Social Impact The social impacts of resource use policies was measured in terms of the net migration of people from country towns each year. Defining the attribute in this way allows for different levels of depopulation to be specified for alternative resource use policies. However, it does not allow for a net increase in population. The accommodation of population growth would add considerable complexity to the analysis of the choice data. Responsibilities and funding mechanism.
Another role of the focus group work was to identify possible mechanisms for funding environmental programs that could act as a payment vehicle and to gauge community sentiment about the notions of environmental responsibilities so that payment vehicle bias could be minimised. The discussions revealed that: •
In the main, participants believed that it was society’s responsibility to pay for programs that addressed land and water degradation. It was accepted that farmers should not be held accountable for all the mistakes of the past.
•
Despite the acceptance of this principle, “free-riding” behaviour was exhibited in many groups. In other words, participants supported the principle of spreading the costs across different sectors of society, providing they did not have to pay anything personally.
•
There was support for the concept of an environmental levy. Participants were familiar with this funding mechanism owing to the various examples of these types of specialpurpose levies (Gun Buy-Back, East Timor, Medicare). Furthermore, just prior to the focus group meetings, there was a considerable amount of debate in the West Australian and New South Wales media about the possible introduction of a salinity levy.
17
•
Notwithstanding the “in principle” support for a levy, there was a general distrust of government. In particular, participants did not trust governments to manage the funds and spend them wisely. They believed that existing tax revenue is being wasted. Across all groups, there was a strong demand for information about the mechanics of how an environmental fund would be managed. Many participants said they would discard the questionnaire if it did not outline how the payment scheme was to be implemented and managed.
•
Participants were very concerned about the equity implications of imposing a tax-based levy. They wanted to know whether the levy would be means tested and whether it would cause financial hardship to the disadvantaged.
•
The city-based focus group participants found it implausible that a special on-going levy would be introduced to fund environmental projects in just one region of the State (particularly evident in the Brisbane group).
These sentiments were taken into account when designing the questionnaire. In an evolution to previous Australian applications of stated preference surveys, greater attention was paid to describing the features of the proposed levy scheme. Respondents were told that a trust fund would be established and managed by a committee independent of government (see Appendix C for details). 4.3 Development of choice options The valuation exercise was introduced to respondents by explaining that public money is currently being spent on a wide range of environmental projects and that this level of action will result in a specific set of future outcomes (the business as usual scenario). Respondents were told that extra investment would be required if additional improvements are to be achieved. An environmental levy on households was proposed as a means of funding this extra action. The questionnaire introduced the concept of a household levy to be paid each year for the next 20 years. A specific level of payment was associated with each choice option, being zero for the business as usual scenario and $20 to $200 for the ‘levy’ options. The attribute levels associated with each option, including the business as usual scenario, were expressed relative to a benchmark, namely a ‘do nothing’ scenario. Under this scenario, it is assumed that even the current level of remedial work is not undertaken. Figure 4.1 demonstrates the three scenarios and provides an example of future attribute levels based on the Species Protection attribute. For each of the levy options, the levels of the environmental attributes were stipulated to always increase over time relative to the business as usual scenario. However, for the Social Impact attribute, both positive and negative outcomes were allowed. This takes account of the possibility that some types of environmental programs could displace rural communities (for example, the conversion of farmland into long rotation forestry), and for others to yield a net reduction in the number of people leaving country areas (a positive outcome). The attribute levels were selected from a feasible range of possibilities and systematically combined according to an experimental design. In order to assist respondents with their deliberations, approximations of the current levels of each attribute were summarised in the introduction to the questionnaire (see Tables 4.2, 4.3, and 4.4 for details).
18
Figure 4.1: An example of a scenario outcomes for the level of endangered species. Note the chart is not drawn to scale.
No. endangered species Do nothing
800 50 protected
Business as usual
140 protected
750
Levy option
660
560
2000
Year
2020
19
Table 4.2: Attribute levels for the national questionnaire Attribute
Current level
Information source for current level.
Business as usual funding
Range 1
Range 2
Range 3
50 protected
70
140
200
1000km restored
5000
8000
10,000
(2020 levels) Species
560 endangered
State of the Environment Report, 1996. pp 4-34 Australian National Parks and Wildlife Service (1992), published in the “Australian National Strategy for Conservation of Species and Communities Threatened with Extinction”. Estimate does not include vulnerable and threatened species.
Waterway health
15,000km degraded
State of the Environment Report, 1996. pp 4-26
Look of land
12 mill ha degraded or unprotected
Science, Engineering, and Innovation Council (1999) published in “Moving Forward in Natural Resource Management”, p. 13.
4 mill ha rehabilitated
6 mill
8 mill
10 mill
Country communities
8000 people leaving annually
ABS Catalogue 3218.0. Estimate based on the 20 Statistical Local Areas in Australia that suffered the highest decline in population in 1998/99.
15000
5000
10,000
20,000
Levy
$0
0
20
50
200
30% of waterways are estimated to be in extremely poor condition (Managing Natural Resources in Rural Australia for a Sustainable Future, 1999).
20
Table 4.3: Attribute levels for the Fitzroy Basin questionnaire Attribute
Current level
Information source for current level
Business as usual funding
Range 1
Range 2
Range 3
(2020 levels) Species
20 endangered
Central Queensland Strategy for Sustainablity, (1998). Only includes vascular plants and fauna.
5 protected
10
15
20
Waterway health
1000km degraded
Queensland State of the Environment Report, 1999. p. 7.42
100 restored
500
800
1000
Look of land
1 mill ha degraded or unprotected
Estimate refers to the area of remnant vegetation on private land that remains unprotected, plus areas affected by soil erosion.
250,000 protected
500,000
750,000
1mill
Country communities
450 people leaving annually
ABS Catalogue 3218. Calculated by summing the population loss in 1998/99 across all Statistical Local Areas in the Fitzroy Statistical Division.
1200
450
1000
1500
Levy
$0
0
20
50
200
21
Table 4.4: Attribute levels for the Great Southern questionnaire Attribute
Current level
Information source for current level
Business as usual funding
Range 1
Range 2
Range 3
(2020 levels) Species
120 endangered
WA Department of Conservation and Land Management, published in Western Australian Salinity Strategy (2000).
25 protected
35
70
100
Waterway health
800 km degraded
Western Australian Salinity Strategy (2000).
100 km restored
250
500
800
Look of land
1 mill ha degraded or unprotected
Approximately 0.5 mill hectares is salt-affected land (Western Australian Salinity Strategy, 2000).
250,000ha rehabilitated
500,000 mill
750,000
1mill
Country communities
520 people leaving annually
ABS Catalogue 3218. Calculated by summing the population loss in 1998/99 across all Statistical Local Areas in the Upper and Lower Great Southern Statistical Divisions.
1500
500
1200
2000
Levy
$0
0
20
50
200
The other 0.5 million constitutes eroded land and unprotected remnant vegetation on private property.
22
4.4 Choice set design The choice sets were designed to minimise the cognitive burden on respondents and to fulfil the technical requirements of the analysis. As part of the design process, five of the focus groups were asked to assess alternative formats for the choice sets. The main design features that were investigated included: •
the presentation of attribute levels in marginal or absolute terms;
•
the presentation of choice options as either generic or labelled alternatives;
•
the presentation of attribute levels in numerical format or the use of icons; and
•
the presentation of choice options in columns or horizontal rows.
Marginal versus absolute format
In the absolute format, attribute levels were expressed relative to a ‘do nothing’ scenario, in which not even the current level of investment is undertaken. For example, respondents were told that the number of endangered species protected under the current level of funding will be 0 + x, while the levy option will protect 0 + x + y species. In the marginal format, respondents were presented only with improvements that are additional to what would be achieved under the existing level of funding. Hence, business as usual outcomes were set to zero and the levy options were set to 0 + y. An example for the Species attribute is shown in Table 4.5. The focus groups showed a clear preference for the choice set in which attribute levels were presented in absolute terms, using the do nothing option as a base. The marginal format was confusing to some people in the focus group studies, so it was rejected in favour of the absolute format. Table 4.5: Presentation of attribute levels using two alternative formats Number of species protected Scenario
Absolute format
Marginal format
Do nothing
0
-
Business as usual
50
0
Levy
140
90
Option labels versus generic options
A choice set with generic options refers to the situation where each option is only described in terms of an attribute profile, which consists of a specified combination of attribute levels. Options are differentiated with a simple nomenclature such as Option A, Option B etc. In contrast, a labelled choice set refers to the situation where each option is given a policy label. The label describes the type of policy or mechanism that would be used to produce the attribute outcomes. Essentially, the label provides the respondent with an additional piece of information upon which to base his/her choice.
23
Figure 4.2: Example choice set:
Members of the focus groups were shown both types of choice sets, labelled and generic. Most people preferred the labelled options as they found it easier to choose between the options. Respondents liked the labels because they provided information about the programs or mechanisms that were driving the outcomes. However, in spite of this demand for labels, it was decided to retain the generic options format. This decision was made because previous research has shown that labels can prompt respondents to trivialise the attributes when making their choice, thereby reducing the statistical explanatory power of the attributes in the choice model (Blamey et. al. 1999). Clearly, this would have been undesirable for this Project where the objective was to estimate attribute values for the purposes of benefit transfer.
24
The request by focus groups for policy labels was addressed by including a statement in the survey introduction. The statement emphasised that many types of projects could be undertaken to improve the environment and viability of country communities, and that different combinations of projects would lead to different outcomes. Respondent were then asked to choose between the options on the basis of attribute outcomes. Attribute icons
Most CM applications have relied entirely on numerical values to convey information about attributes and their levels. This presentation format may be confusing to some respondents and cause fatigue. In this study, visual stimuli were incorporated into the choice sets in an effort to improve respondent cognition and promote interest. An icon was used to represent each attribute and the size of the icons was scaled to denote the level of the attribute. Figure 4.2 contains an example of a choice set. 4.5 Pre-testing The survey instrument was pre-tested over two days in suburban Sydney using a door-to-door, drop off and pick up method. The suburbs selected for the pre-test contained households from a broad range of socioeconomic groups. In total, 25 households were interviewed. Only minor modifications were made to the questionnaire following the pre-testing phase as debriefs with the respondent households did not reveal any significant communication problems. 4.6 Sampling A market research firm (Barbara Davis and Associates) was contracted to draw random samples from “Australia on Disk,” a telephone directory database of the Australian population. The size of the total sample was 10,800 households. Table 4.6 contains a breakdown of the population sub-samples. Table 4.6: Size of sub-samples, by questionnaire version Questionnaire Version National
Great Southern
Fitzroy
National
3200
-
-
Albany
1200
1200
-
Rockhampton
1200
-
1200
Perth
-
1400
-
Brisbane
-
-
1400
Population samples
4.7 Survey administration Barbara Davis and Associates was also engaged to administer the survey. The questionnaires were mailed out to households with a covering letter outlining the objectives of the survey. No incentives were provided as a means of increasing response rate.
25
Respondents were asked to use the reply-paid envelope provided to return their completed questionnaire. Households who failed to return a questionnaire within two weeks were sent a reminder notice. A second reminder was sent out after four weeks had elapsed from the time of the first mail-out. The questionnaire was in the field for a period of approximately six weeks. At the end of the survey period, a follow-up telephone survey of non-respondents was conducted. The purpose of this survey was to identify the reasons why households did not respond and to determine whether non-respondents had significantly different characteristics to respondents.
26
Chapter 5: Descriptive overview of survey results This chapter contains an overview of the survey results. The purpose of the overview is to provide an initial description of the data to assist with interpreting the model results. In Section 5.1 the response rate to the survey is reported for each population sub-sample. Particular attention is paid to the proportion of respondents who completed the choice task. Section 5.2 contains a description of the data. Key characteristics of the national sample are summarised and compared against census statistics to determine the representativeness of the sample. Section 5.3 presents the results of a preliminary assessment of respondents’ willingness to pay an environmental levy. The last section of this chapter (5.4) contains details of the choice model specifications and reports the parameter estimates for the model. An assessment is made of the statistical significance of the models and the extent to which the model coefficients accord with theoretical expectations. 5.1 Response rate Overall response
The overall response rate to the survey was 16 per cent which equated to 1569 completed questionnaires (Figure 5.1). This response rate is net of the 10 per cent of questionnaires that were undeliverable due to outdated address details. Of those respondents who completed a questionnaire, the majority (89 per cent) answered all five choice questions, while a small proportion (8 per cent) only answered a subset of the five questions. Three per cent of respondents failed to complete any of the choice questions. Figure 5.1: Response rate
Total mailout 10,800
Undeliverable 1079 (10%)
Completed 1569 (16%)
All choice questions answered (89%)
Some choice questions answered (8%)
Delivered 9721 (90%)
No response 8152 (84%)
No choice questions answered (3%)
27
There were significant differences in response rate across the samples2. Table 5.1 contains a summary of the response rates by sample and type of questionnaire version administered. The main points to note are: •
The lowest response rate was from Brisbane (13 per cent). In contrast, the Perth response rate was 18 per cent.
•
The response by the regional samples (Albany and Rockhampton) to the case study questionnaires is not significantly greater than the response by these same samples to the national questionnaire. Response rates to both versions of the questionnaire range from 14 to 17 per cent. The variation in response between Albany and Rockhampton is not statistically significant at the five per cent level.
•
There is no statistical difference in response rates between metropolitan and nonmetropolitan residents.
Table 5.1: Response rates for each sample, by questionnaire version. Questionnaire version Sample
National
Fitzroy
Grt Southern
Metropolitan
18%
-
-
Non-metropolitan
17%
-
-
Perth
-
-
18%
Brisbane
-
13%
-
Albany
17%
-
16%
Rockhampton
14%
16%
-
16%
15%
17%
National
Capital city
Regional
State differences in response rate.
Within the national sample there is a large degree of variation in response rate across the States (Table 5.2). Owing to the small sample size for some States, not all the differences are statistically significant. However, ACT’s response rate is significantly higher than that of NSW and WA. The education levels of respondents and their environmental disposition are reported in Table 5.3. There is no evidence of a statistically significant correlation between these factors and response rate.
2
The differences are statistically significant at the 5% level using a chi-squared test.
28
Table 5.2: Response rate for the national sample, by State and Territory Total mailout
Delivered
Completed
Response rate
NT
20
16
2
13%
ACT
54
47
13
28%a
TAS
73
67
11
16%
SA
266
246
50
20%
WA
307
264
35
13%
Qld.
592
534
95
18%
Vic.
800
719
131
18%
NSW
1088
944
153
16%
3200
2837
490
17%
a
Significantly different from NSW and WA at the 5% level, using a chi-squared test.
Table 5.3: Education level and environmental disposition of respondents, by State and Territory. The sub-sample containing NT respondents is excluded owing to its small sample size. Proportion of respondents who..... Response rate
hold a tertiary degree
support an environmental organisation
ACT
28%
54%
31%
TAS
16%
36%
36%
SA
20%
38%
30%
WA
13%
49%
29%
Qld.
18%
27%
18%
Vic.
18%
37%
16%
NSW
16%
33%
29%
Table 5.4: Choice set completion rate by age group and education level. Under 55
a
55 and over
All ages
Primary
80%
81%
81%
Yr 10
88%
83%
85%
Yr 12
93%
84%
90%
Diploma
92%
89%
91%
97%
92%
95%
93%
86%
Tertiary All levels
b
a
a
Variation in completion rate across education level is significant at the 5% level, using a chi-squared test. b The difference in completion rate between the two age groups is significant at the 5% level, using a chi-squared test.
29
5.2 Completion of the choice task. Of those responding to the questionnaire, 11 per cent failed to complete all or some of the choice questions. The results in Table 5.4 suggest that education and age are significant determinants of respondents choosing to ignore or only partially complete the choice tasks. The values in the table are choice set completion rates, calculated as the proportion of respondents who returned a questionnaire and completed all the choice questions. There is a statistically significant increase in the completion rate with progressively higher education levels, and this is most noticeable for respondents aged 55 years or under (the effect of education is not significant for respondents in the older age group). Completion rate is significantly lower for respondents over the age of 55. In order to discover what other factors are important in determining completion rates, respondent reactions to the questionnaire were analysed for two groups of participants: Those who completed all the choice questions and those who did not. It was found that a significantly larger proportion of respondents in the latter group found the background information confusing and fewer felt they needed more information (Table 5.5). It would appear that a small percentage of respondents, mostly those with low education levels, had difficulty understanding the issues and trade-offs that were being presented to them. 5.3 Description of data Sample characteristics
A summary of the key socioeconomic characteristics for each of the five samples is contained in Table 5.6. Albany and Rockhampton stand out because they both contain the highest proportion of respondents in the low-income bracket. The proportion of respondents with pro-environment sentiment differs considerably across the samples, ranging from 13 per cent for Rockhampton up to 27 per cent for Albany. The survey appears to have been self-selecting for male respondents, particularly in the metropolitan city samples. Sample selection bias is discussed at greater length in the following section which examines the representativeness of the national sample. Some of the socioeconomic characteristics used to describe respondents are weakly correlated with each other. The notable positive correlations are between education and income, and between age and sex (the probability of a respondent being male increases with age). The correlation coefficient for “green” disposition and income is positive but the correlation is not significant. Among the negative relationships, only the correlations between age, income, and education level are significant. A full correlation matrix for all the socioeconomic variables is contained in Table 5.7.
30
Table 5.5: Reactions to the questionnaire by two groups of respondents: Those that completed all choice questions and those who did not. The proportion of respondents who answered “YES” to the statement are indicated. Choice questions completed
Choice questions not completed
I needed more information
32%
24%a
I thought the information was biased
21%
22%
I thought the information was confusing
16%
26%a
I found options confusing
28%
32%
I thought the options were unrealistic
18%
24%
I think that a levy will one day be introduced
60%
51%a
a
Difference in completion rate between the two groups is significant at the 5% level using a chisquared test.
Table 5.6: Selected socioeconomic characteristics of the samples. National
Perth
Brisbane
Albany
Rock’n
Modal income category
$36,40051,999
$52,00077,999
$36,40051,999
$623915,599
$623915,599
Modal education category (highest qualification)
Tertiary degree
Tertiary degree
Tertiary degree
Diploma / certificate
Tertiary degree
Modal age group
45-54
45-54
35-44
65 +
35-44
% supporting green group(s)
24%
22%
22%
27%
13%
1.6 to 1
1.5 to 1
1.8 to 1
1.3 to 1
1.3 to1
490
217
170
356
336
Male to female ratio Sample size
Table 5.7: Correlation matrix for the socioeconomic variables Sex Sex
Citizen
Green
Education
a
0.1735
1.0000
-0.0029
-0.0297
1.0000
Greenb
-0.0945
-0.0756
0.0046
1.0000
Education
-0.0174
-0.2590
0.0186
0.1608
1.0000
Income
0.0651
-0.2836
-0.0119
0.0869
0.3095
a
Income
1.0000
Age Citizen
Age
Australian citizenship; organisation.
b
1.0000
Indicator of whether respondent is a member of, or donates to, an environmental
31
Representativeness of the national sample
The sample of people who responded to the national survey is not representative of the Australian population with respect to some key socioeconomic characteristics. Notably, the male-to-female ratio of respondents is disproportionately large relative to the national average, which suggests that males were more likely to complete the questionnaire. Further evidence of sampling bias is apparent when the sample statistics are compared alongside the national census data: •
Younger age groups are under-represented (Figure 5.2).
•
The sample contains a disproportionately large group of high-income earners (Figure 5.3).
•
35 per cent of respondents have a tertiary degree which is more than double the national level of 14 per cent (Figure 5.4).
•
24 per cent of respondents reported that they donated to, or were members of, an environmental organisation. There is evidence to suggest that this level of commitment to environmental causes exceeds the national average. Whilst directly comparable statistics are not available, the Australian Bureau of Statistics has estimated that only nine per cent of Australians rank environmental problems as their top social issue (ABS, 1999). The Australian Conservation Foundation estimates that five per cent of the national population belong to at least one environmental organisation (M. Fogarty pers. comm. 2000).
Figure 5.2: Age composition of the national sample relative to ABS estimates for the Australian population aged 18 years and over (Catalogue 3201, 1999). 65 and over
55-64
45-54
35-44
National sample
25-34
Australian population
18-24
0%
5%
10%
15%
20%
25%
30%
32
Figure 5.3: Income distribution for the national sample relative to ABS estimates for the Australian population (all income units, Catalogue 6253, 1999). Over 104000 78000-103999 52000-77999 36400-51999 26000-36399 15600-25999 6239-15599
National sample Australian population
Under 6239 0%
5%
10%
15%
20%
25%
30%
Figure 5.4: Highest level of education attained by respondents relative to the Australian population (ABS, 1998, Catalogue 4224).
Tertiary degree
Diploma or trade
Year 12
National sample Australian population
Primary and Year 10
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
33
Figure 5.5: Proportional representation of respondents across the States compared to the 1998 distribution of the Australian population aged 18 years and over (ABS Catalogue 3201). Response rates for each State are shown by the labels alongside each bar. NSW
16%
VIC
18%
QLD
18%
WA
13%
SA
20%
TAS
16%
ACT
28%
National sample NT 0.0%
Australian population
13% 5.0%
10.0%
15.0%
20.0%
25.0%
30.0%
35.0%
40.0%
While the national sample is not representative of the Australian population for a number of important socioeconomic characteristics, it does contain a satisfactory representation of respondents from each State. The proportion of respondents from each State is approximately equivalent to the geographic distribution of the Australian population, with the main exceptions being WA, NSW and the NT which are slightly under-represented (Figure 5.6). The poor response from WA and the NT (13 per cent) is partly responsible for the underrepresentation in these States. The ratio of respondents from capital cities and non-metropolitan areas is approximately 2:1. This figure is higher than the ratio published by the Australian Bureau of Statistics (Catalogue 3222.0), which is 1.8:1. The proportion of non-Australian citizens in the sample is four per cent which is below the eight per cent of Australian residents who are estimated to be citizens of an overseas country (ABS, Catalogue 3412.0). These statistics suggest a sampling bias towards metropolitan residents and against the inclusion of people without Australian citizenship. 5.4 Preliminary assessment of willingness to pay Of those respondents who completed all the choice questions, 20 per cent consistently selected the business as usual option in each choice set. The other 80 per cent of respondents chose at least one of the options that involved a levy. This proportion of respondents in favour of a levy exceeds the estimate obtained in a survey of Western Australian households conducted by Patterson Market Research in December 1999. A telephone poll of 400 households revealed that 55 per cent of respondents were willing to pay a levy dedicated to addressing this State’s salinity problem. These conflicting results provide further evidence to suggest that the present study self-selected for pro-environment respondents. Table 5.8 contains a detailed break-down of those respondents who chose the business as usual option, by sample and questionnaire version. This analysis shows that the metropolitan sample 34
issued with the national questionnaire contains the lowest proportion of respondents selecting the status quo (15 per cent), while the Rockhampton sample issued with the same questionnaire has the highest proportion (24 per cent). The proportion of status quo responses by the non-metropolitan sub-sample lies in between these two extremes. The differences provide preliminary support for the hypothesis that values are variable across different population groups and questionnaire frames. This initial review of the data suggests that nonmetropolitan respondents, particularly the Rockhampton sample, have lower values than their city-based counterparts. However, there are numerous reasons why respondents may be unwilling to select a levy option. Some may have a genuine low value for the environment and country communities, while others could be trying to influence the results of the survey by protesting against a levy. Another possibility could be that respondents are distrustful of the government and have misgivings about the efficiency with which the funds will be spent. In order to investigate what factors were primarily responsible for people opting not to pay a levy, respondents who consistently selected the business as usual option were asked to tick off the most important reason influencing their choice. A summary of their responses is contained in Table 5.9. The key findings of this analysis are: •
The dominant reason given for rejecting the environmental levy was that the levy was not affordable. Twenty to 30 per cent of respondents are in this category. A separate crosstabulation reveals that most of the people in this category have incomes that are below the sample average. Consequently, it can be concluded that the zero bids given by these respondents are likely be “true” zeros rather than protests.
•
Another reason given for selecting the status quo was opposition to the levy. This response is highest among Albany, Rockhampton and Brisbane respondents (20 to 30 per cent) but significantly lower opposition was recorded for National and Perth respondents (10-11per cent).
•
Distrust of the government was ranked as a primary reason by 6 to 14 per cent of respondents. If these respondents are added to those who stated their opposition to the levy, then the Queensland samples contain the highest proportion of respondents with “protest” bids (approximately 35 per cent).
•
Ten to 11 per cent of respondents believed that land and water resources were already well managed and cited this as their main reason for rejecting a levy.
35
Table 5.8: Proportion of respondents who selected the status quo option for all choice questions. Questionnaire version Sample
National
Fitzroy
Grt Southern
Metropolitan
15%a
-
-
Non-metropolitan
22%
-
-
Perth
-
-
18%
Brisbane
-
22%
-
Albany
23%
-
18%b
Rockhampton
24%
19%
-
20%
18%
21%
National
Capital city
Regional
a a
Significantly lower than the non-metropolitan sample at 5% probability level. Significantly lower than the Albany-National sample at 5% probability level.
Table 5.9: Nominated primary reason for selecting the business as usual option, by population sample. Values are the percentage of respondents who nominated the stated reason as their primary motivation. National
Albany
Perth
Rock’n
Brisbane
Land and water already well managed
9%
11%
3%
6%
0%
Cannot afford the levy
31%
30%
32%
25%
17%
Oppose the levy
10%
19%
11%
21%
29%
Distrust the government
14%
6%
14%
13%
6%
Did not know which option was best, so stuck with the status quo.
4%
8%
3%
0%
3%
Other reason
14%
16%
27%
25%
31%
No response or multiple reasons given.
17%
10%
11%
10%
14%
77
63
37
63
35
Total number selecting status quo
5.5 Model specification and parameter estimates Specification
A nested structure was used to model respondents’ choices of alternative options3. This structure assumes that respondents made an initial decision to either support an environmental levy or go with the status quo option (Figure 5.6). If the levy was supported, then the respondent was faced with a second-level decision that involved the choice between two different levy options (B and C). This lower-level decision was “nested” below the initial decision. The two levels of the nest are linked by an “inclusive value” which embodies the
3
Initially a multinomial logit model was used to describe the data relationships. However, this specification was shown to result in breaches of the Independence of Irrelevant Alternatives (IIA) assumption.
36
expected utility associated with the lower-level decisions. The inclusive value is included as a variable in upper-level utility functions. In this study the upper-level decision was hypothesised to be influenced by the respondent’s socioeconomic characteristics (age, sex, income), environmental disposition, and whether or not the respondent was confused by the background information4. The probability of the levy being supported was expected to increase with income and pro-environment sentiment, but decrease for respondents who reported confusion. In addition to these individual-specific variables, the choice between retaining the status quo or paying a levy was assumed to be influenced by the expected utility (or inclusive value) associated with each alternative and a constant term for the levy alternative. The lower level decision between the alternative levy options was hypothesised to be influenced by the attributes of each option. A technical summary of the model specification is contained in Box 5.1 and the variables are described in Table 5.10. Figure 5.6: Diagram of the nesting structure adopted for the choice model Support for proposal
Status Quo (Option A)
Levy
Option B
Option C
Box 5.1: Specification of the utility functions. The upper level utility functions of the nested logit model were specified as follows: Vlevy = ASC + β1Sex + β2Age + β3Income + β4Green + β5Confuse + α1IVlevy VSQ = α2IVSQ where Vlevy is the utility associated with the levy options and VSQ is the utility obtained from selecting the status quo option. The alternative specific constant (ASC) is specified for the levy option, and the socioeconomic characteristics are incorporated into the model as interactions with this ASC. The IV variables are inclusive values from the lower level of the nest. The coefficient on the inclusive value for the status quo option (α2) is fixed to one because only one alternative exists in the lower level nest for this option. The utility functions for each of the lower-level choice options are specified in terms of attributes. The utility for option j is given by: Vj = β6Species + β7Look + β8Water + β9Social + β10Cost where j is option A (the status quo), B, or C.
4
Missing observations for respondent characteristics were replaced with modal values for the sample.
37
Table 5.10: Description of variables used in the choice models. Variable
Description
Species
Endangered species, measured by the number of species protected from extinction.
Look
Landscape aesthetics, measured by the area of farmland repaired and bush protected (hectares).
Water
Waterway health, measured by the total length of waterways restored for fishing or swimming (kilometres).
Social
Viability of country communities, measured by the net annual loss of population from country towns.
Cost
The environmental levy, measured as an annual levy on household income
ASC
Alternative specific constant for the levy option, assigned a value of 1 for options B and C and zero otherwise.
Sex
Respondent’s gender, assigned a value of 0 for females and 1 for males.
Age
Respondent’s age category, ranging from 1 to 6 (youngest to oldest).
Income
Respondent’s before-tax household income category, ranging from 1 to 8 (lowest to highest).
Green
Dummy variable assigned a value of 1 for respondents who are members of, or donate to, an environmental organisation and 0 otherwise.
Confuse
Dummy variable assigned a value of 1 for respondents who reported that they found the background information confusing, 0 otherwise.
IV
Inclusive value representing the expected utility from alternatives in the lower level of the nest.
Parameter estimates
Seven nested logit models were estimated, each model being specific for a combination of questionnaire version and population sample. A summary of parameter estimates and their statistical significance is contained in Table 5.11. The models exhibit a satisfactory goodness of fit with Likelihood Ratio Indices (LRI) ranging between 0.17 and 0.26. Parameter estimates for the attributes conform to a priori expectations. For the majority of models estimated, the environmental attributes (Species, Look, and Water) are statistically significant and have positive signs, which indicates that increases in the levels of these attributes add to an individual’s utility. One exception to this conclusion is Species in the Fitzroy models, which is not significant in either of these models. This suggests that the protection of Species is not perceived to be a priority issue in the Fitzroy Basin. The only other exception is Water in the Albany-National model. The results suggest that Albany respondents do not perceive this attribute to be important in the national context, although, at the local level, it is highly significant (Model 4). The signs on Social and Cost are significant and negative across all models, which means that utility is reduced by increases in the levy and higher levels of population loss from country areas. The individual-specific socio-demographic variables (Sex, Age, Income, Green, and Confuse) are also significant in explaining respondent choices. The probability of choosing a levy option is shown, in most models, to increase with a respondent’s income and pro-environmental disposition. This finding supports the validity of the models, as willingness to pay should be
38
underpinned by an ability to pay. Perth was the only sample for which the choice of levy was independent of income. Confuse is a significant variable in all but one of the models. Its negative sign agrees with the prior that respondents who were confused by the questionnaire were more inclined to choose the status quo option. Age and Sex are significant in some of the models but the effect of these variables on choice is not consistent. In several of the models age has a negative sign which implies that older respondents selected the status quo in preference to a levy.
39
Table 5.11: Parameter estimates for the nested logit choice models. Each model is specific for a population sample and questionnaire frame Model
1
2
3
4
5
6
7
Frame
National
National
National
Great Southern
Fitzroy Basin
Great Southern
Fitzroy Basin
Population
National
Albany
Rockhampton
Albany
Rockhampton
Perth
Brisbane
Lower level choice variables SPECIES
5.49E-03 **
2.39E-03 *
2.89E-03 *
1.28E-02 **
4.07E-03
1.13E-02 **
1.72E-02
LOOK
6.01E-08 **
1.84E-07 **
2.04E-07 **
1.52E-06 **
8.07E-07 **
1.24E-06 **
1.11E-06 **
WATER
6.33E-05 **
4.55E-05
7.54E-05 **
1.29E-03 **
1.04E-03 **
8.05E-04 **
6.71E-04 **
SOCIAL
-6.94E-05 **
-9.46E-05 **
-6.74E-05 **
-4.52E-04 **
-1.15E-03 **
-6.34E-04 **
-8.78E-04 **
COST
-8.13E-03 **
-8.78E-03 **
-1.04E-02 **
-8.28E-03 **
-5.14E-03 **
-8.89E-03 **
-8.54E-03 **
-1.00E+00 **
2.40E+00 **
-2.02E+00 **
9.30E-01 **
2.54E+00 **
2.39E+00 **
Upper level choice variables ASC
-5.85E-01 **
SEX
-3.24E-01 **
5.01E-01 **
-5.96E-01 **
5.70E-01 **
-6.94E-01 **
-2.43E-01
-2.89E-01 *
AGE
7.96E-02 **
-1.22E-01 **
-3.50E-01 **
9.03E-02
-7.39E-02
-3.83E-01 **
-4.47E-01 **
INCOME
2.62E-01 **
2.13E-01 **
1.72E-01 **
GREEN
2.47E-01 **
4.50E-01 **
6.49E-01 *
1.31E+00 **
-7.07E-01 **
-6.77E-01 **
-1.05E+00 **
-7.74E-01 **
CONFUSE
3.48E-01 **
1.15E-01 **
-5.71E-03
2.02E-01
-1.39E-01
-6.37E-01 **
9.65E-02 ** -3.22E-01
-3.62E-01 *
Inclusive values IV staus quo IV levy
1 0.3434 **
1 0.3914 **
1 0.1950
1 0.2461 *
1 0.2262
1 0.3595 **
1 0.0618
No choice sets
2329
860
720
765
818
1046
823
Log Likelihood
-2196.05
-803.75
-645.29
-683.77
-802.10
-976.78
-761.39
LRI
0.2271
0.2155
0.2419
0.2698
0.1770
0.2337
0.2302
LRI adjusted
0.2251
0.2099
0.2355
0.2641
0.1709
0.2293
0.2251
Notes: * denotes significance of parameter at the 10% level, ** denotes significance at the 5% level.
40
Chapter 6: Value estimates from the national survey 6.1 Overview The results reported in this chapter of the report relate to the national questionnaire in which respondents were asked to make choices between policy outcomes that have an impact at a national level. Two types of value estimates are provided: Implicit prices and welfare impacts (see Box 6.1 for details on how these estimates are calculated). Attribute implicit prices are a measure of the willingness of respondents to trade-off household income to secure a single unit increase in a particular environmental or social attribute. Implicit price estimates are most useful when assessing the non-market impact of policies that have single-attribute outcomes. If a management policy is expected to affect the levels of multiple attributes, then an approximation of the benefit generated can be obtained by aggregating the implicit prices of all the attributes affected. However in such circumstances, particularly when the changes in attributes are relatively large, more accurate estimates of changes in welfare can be achieved using the full choice model. This welfare measure is known as ‘compensating surplus’ and represents the total value of a change in the levels of multiple attributes away from the business as usual scenario. Use of the full choice model incorporates the impacts of the attributes, as well as the factors influencing choice that have not been defined in the choice sets. In other words, the implicit prices of the attributes alone do not account for the total welfare impact. 6.2 Attribute implicit prices Implicit price estimates for each of the attributes are summarised in Table 6.1. The estimates are a measure of the amount that households are willing to pay each year for the next 20 years to secure an environmental or social improvement. Across both regional and national samples, respondents hold positive values for environmental attributes, whilst negative implicit prices are estimated for losses of people from country communities. This result implies that respondents perceive depopulation as a cost and are willing to trade-off income to prevent a loss in community viability. For the national sample, respondent households are willing to pay, on average, 68 cents per annum over the next 20 years for every species that is protected from extinction. The value of Landscape Aesthetics is estimated to be 7 cents per 10,000 hectares of bushland protected or farmland restored, while a similar amount (8 cents) is estimated to be the value for every 10 kilometres of waterway restored. A negative implicit price of 9 cents is estimated for every 10 people leaving country communities. The implicit price estimates assume non-diminishing values for additional improvements in attribute levels. While a non-linear relationship would be expected, at least beyond a certain level of improvement, transforming the data to allow for non-linearity did not improve the model fit. Therefore, it is concluded that implicit prices are constant for changes in the attributes over the range of levels used in the choice sets.
41
Box 6.1: Implicit prices and welfare calculation The implicit price (IP) for an environmental or social attribute is equivalent to the marginal rate of substitution between the attribute and the levy. Thus, the implicit price for an attribute i is calculated as follows:
IPi =
βi − β COST
The welfare impacts for a change in environmental and/or social outcomes are measured in in terms of compensating surplus (CS). For the nested logit models specified in this study, the calculation is as follows:
CS =
V 1 −V 0 − β COST
where V0 is the utility associated with the status quo option, which is given by: V0 = α2(β6Species + β7Look + β8Water + β9Social) and V1 is the utility associated with the change option, given by: V1 = (ASC + β1Sex + β2Age + β3Income + β4Green + β5Confuse) + α1(β6Species + β7Look + β8Water + β9Social). V0 is calculated using base levels for the attributes, while V1 is calculated using levels associated with the change scenario. Sample modes were used for the socio-economic variables (all of which are categorical).
The values held by respondents from regional areas are of a similar order of magnitude to those of people in the national sample, although some differences are evident. Differences that are statistically significant include: •
Species Protection is more highly valued by the national sample of households compared to the regional samples; and
•
Landscape Aesthetics is more highly valued by regional respondents than the national sample.
Given that the majority of households in the national sample are from metropolitan city areas (68 per cent), these differences could indicate that city dwellers place a higher weighting on Species Protection (a non-use value) relative to country dwellers and a lower weighting on Landscape Aesthetics.
42
Table 6.1: Implicit prices estimated for attributes in the national context Species protection
Landscape Aesthetics
Waterway Health
Social Impact
$ per species protected
$ per 10,000 ha restored
$ per 10 km restored
$ per 10 persons leaving
Lower estimate
0.47
0.02
0.04
-0.11
Best estimate
0.68
0.07
0.08
-0.09
Upper estimate
0.88
0.14
0.16
-0.07
Lower estimate
-0.03
0.14
0.00
-0.14
Best estimate
0.27
0.21
0.00A
-0.11
Upper estimate
0.51
0.29
0.00
-0.08
Lower estimate
0.03
0.12
0.01
-0.09
Best estimate
0.28
0.20
0.07
-0.06
Upper estimate
0.58
0.30
0.14
-0.08
National sample
Albany sample
Rockhampton sample
A This attribute is ‘not statistically significant’ from zero *Best estimate denotes the mean value while the upper and lower estimates represent the 95% confidence interval.
6.3: Welfare impacts from alternative scenarios The choice model derived from the national sample of respondents was used to estimate the welfare impacts (compensating surpluses) of four alternative resource use scenarios. The impacts are measured relative to a fifth scenario; the ‘business as usual’ option. The four change scenarios are indicative of the twenty-year outcomes that could eventuate under alternative management regimes (Table 6.2). This analysis demonstrates how the choice model can be used to estimate the benefits of environmental and/or social improvements (benefits gross of the costs of implementing the changes). Results of the analysis are summarised in Table 6.3 and are described below. Biodiversity protection scenario
This scenario describes the possible outcomes from policies designed to promote biodiversity protection. It is assumed that an additional 100 species would be protected relative to the business as usual outcome, together with an additional one million hectares of improved landscape aesthetics and 200 kilometres of waterway restoration. The annual value of this policy is estimated to range from $88 to $142 with a best estimate of $112 per annum for 20 years. Expressed as a lump sum present value, the best estimate is equivalent to a one off payment of $1,466 (assumes a 5 per cent discount rate).
43
Table 6.2: Four hypothetical scenarios developed to demonstrate ways that the choice model could be used to estimate the welfare impacts of changes away from the business as usual scenario Attributes
Business as usual Scenario
Biodiversity Protection Scenario
Waterway Restoration Scenario
Negative social impacts scenario
Positive social impacts scenario
Species Protection (Number of species protected)
50
150
75
100
100
Landscape Aesthetics
4 mill.
5 mill.
4.5 mill
6 mill
6 mill
1,000
1,200
5,000
2,500
2,500
15,000
15,000
15,000
20,000
5,000
(Hectares of farmland repaired and bushland protected) Waterway health (Kilometres of waterways restored for swimming and fishing) Social impact (No. of people leaving country areas per year.) Waterway restoration scenario
This scenario involves policies that focus on restoring waterways. It is assumed that an additional 4,000 kilometres of waterways would be rehabilitated by 2020 relative to the business as usual scenario. More modest improvements are assumed for landscape amenity and species protection. Respondent households are estimated to be willing to pay $104 per year for 20 years for the outcomes of this policy, which equates to a lump sum present value of $1,361. Negative social impacts scenario
This scenario involves improvements to all environmental attributes and does not target a particular environmental outcome. However, the policies used to achieve these environmental improvements are assumed to lead to an additional 5,000 people leaving country communities each year relative to the business as usual scenario. Such a scenario could be encountered if trade-offs exist between conservation objectives and regional development. The welfare impact of this scenario is estimated to be $92 per annum per respondent household, which equates to a lump sum present value of $1,204 per household. Positive social impacts scenario
This scenario consists of a set of policies that deliver both environmental and social improvements relative to the business as usual scenario. It is assumed that the number of people leaving country areas is reduced by 10,000 per year so that only 5,000 rather than 15,000 people leave per year. Measured against the business as usual scenario, this is a gain of 10,000 people per year. This outcome could eventuate if conservation management policies were adopted that stimulated regional employment. Households would be willing to pay $136 per annum for 20 years for such an outcome, or $1,780 per household when expressed as a lump sum.
44
Table 6.3: Estimated welfare impacts per household for each of the four hypothetical scenarios * Biodiversity Waterway Negative social Positive social Protection Restoration impacts impacts Scenario Scenario scenario scenario Estimated annual welfare gain per household* Low estimate
$88
$77
$63
$114
Best estimate
$112
$104
$92
$136
Upper estimate
$142
$136
$128
$164
Estimated mean lump sum present value per household
A
Low estimate
(@3%)
$1,348
$1,180
$965
$1,747
Best estimate
(@3%)
$1,716
$1,594
$1,410
$2,084
Upper estimate (@3%)
$2,176
$2,084
$1,961
$2,513
Low estimate
(@5%)
$1,152
$1,008
$824
$1,492
Best estimate
(@5%)
$1,466
$1,361
$1,204
$1,780
Upper estimate (@5%)
$1,858
$1,780
$1,675
$2,146
Low estimate
(@6%)
$1,070
$936
$766
$1,386
Best estimate
(@6%)
$1,362
$1,264
$1,119
$1,654
Upper estimate (@6%)
$1,726
$1,654
$1,556
$1,994
* Estimates derived using a full choice model not the simple multiplication of attribute values A
Discount rates shown in parenthesis
6.4 Variability of values across different household groups Socioeconomic characteristics
Welfare impacts are found to vary substantially over different segments of the Australian community. The ‘negative social impacts’ scenario is used as an example to demonstrate this variability. The analysis was undertaken by varying independently the level of each respondent characteristic in the choice model and recalculating the welfare impact. Table 6.4 contains a summary of estimated welfare impacts, categorised according to demographic and socioeconomic groupings. The main findings are: •
Respondents with a pro-environment disposition are willing to pay $30 more per annum than other respondents (pro-environment respondents are defined as those who currently donate to, or are members of, an environmental organisation).
•
Females have a significantly higher willingness to pay than males, the difference being in the order of $40 per annum. This finding is consistent with the results of a CM study undertaken in the ACT which estimated environmental values associated with water supply options (Centre for International Economics, 1997).
45
•
Values increase with both age and income. For this particular sample of respondents, age is negatively correlated to income and education level, so a different factor must be influencing older respondents to have higher values. It is possible, for instance, that older respondents have a greater sense of social responsibility. The negative estimate for respondents in the lowest income category should be regarded as a zero value. It means that, on average, a respondent with this level of income has a low to zero value for the scenario.
Table 6.4: Variability of welfare impacts across different socioeconomic groups, evaluated for the ‘negative impacts’ scenario. Socioeconomic group
Annual welfare impact ($/household) Mean
95% confidence interval
Environmental disposition pro-environment
122
83 – 168
not pro-environment
92
63 – 128
25-34
72
37 – 108
35-44
82
51 – 116
45-54
92
63 – 128
55-64
102
68 – 141
65 and over
112
75 – 156
Male
92
63 – 128
Female
132
99 – 170
6239-15,599
-5
-40 – 36
15,600-25,999
28
-2 – 66
26,000-36,399
60
32 – 95
36,400-51,999
92
63 – 128
52,000-77,999
124
94 – 162
78,000-103,999
156
123 – 196
more than 104,000
189
152 – 232
Age-group
Gender
Household income
National versus regional
Statistical tests reveal that the welfare impacts of a resource use change are equivalent across respondents from the national and regional samples once socioeconomic differences are controlled for. This test was performed by estimating separate choice models for each sample (Albany, Rockhampton, and national), then substituting the mean age and income values for the national sample into the Albany and Rockhampton models. This substitution procedure effectively removes any inter-sample variation in welfare impacts that are due to age and income differences. Figure 6.1 shows that once socioeconomic differences are allowed for, there is no statistical difference between the welfare estimates calculated for each sample.
46
Figure 6.1: Annual welfare estimates from the negative impact scenario, evaluated for different population samples
Error bars denote 95% confidence intervals Regional sample socioeconomics
200
National sample socioeconomics
hhold)
150
100
50
0 National
Albany
Rockhampton
-50
-100
Population sample
State differences
Value estimates appear to be consistent for respondents from different States. The two States examined were Queensland and Western Australia (WA). These States were singled out because a survey by the Australian Bureau of Statistics (ABS) indicated that WA residents have a greater awareness of environmental problems than any of the other States, and Queenslanders have the lowest levels of awareness (ABS, Catalogue 4602, 1999). State differences were tested by specifying two dummy variables for ‘place of residence’; one for West Australians and the other for Queenslanders. Neither dummy was significant in the choice model, which suggests that people from these states who responded to the survey have the same preference structure. 6.5 Aggregate welfare impacts of resource use change The ‘negative social impacts’ scenario described above is used to illustrate the process of calculating the aggregate non-market impacts of land and water degradation in Australia. The aggregate impact of this scenario is estimated to be $3.9 billion in present value terms (5 per cent discount rate). This is an estimate of the community’s maximum willingness to pay for the specified set of environmental improvements or, alternatively, the size of benefits foregone if these improvements are not undertaken. The estimate is calculated by extrapolating the per household estimate of $1204 (from Table 6.3) to 45 per cent of the Australian population of 7,185,540 households (ABS, 2000). It is not valid to simply aggregate the value estimates to the entire household population because only 17 per cent of households responded to the questionnaire. A conservative approach to aggregation is to assume that all non-respondents have zero values, thus limiting the extrapolation of benefits to just 17 per cent of the population. However, this would almost certainly be an underestimate of the true aggregate benefits. 47
The aggregation factor of 45 per cent is an estimate derived from a follow-up survey of 75 non-respondent households. This survey revealed that 37 per cent of people indicated an interest in the questionnaire but had been too busy to answer it. Another 32 per cent were interested in the topic but felt that the questions were inappropriate. Only seven per cent of the respondents replied that they had no interest in land and water degradation issues. On the basis of these results it appears reasonable to assume that at least 37 per cent of non-respondents hold non-zero values. If this proportion of non-respondents is added to the 17 per cent of households who responded, the aggregation factor is calculated to be 48 per cent of the total household population ([0.17+(1.00-0.17)*0.37] = 0.48.). A slightly more conservative figure of 45 per cent is adopted for this analysis as a best-bet measure. Table 6.5 summarises the aggregate welfare impacts for each of the four scenarios. Table 6.5: Estimated aggregate welfare impacts for each of four hypothetical scenarios * Biodiversity Protection Scenario
Waterway Restoration Scenario
Negative social impacts scenario
Positive social impacts scenario
Estimated lump sum present values (billions) Low estimate
(@3%)
$4.36
$3.81
$3.12
$5.65
Best estimate
(@3%)
$5.55
$5.15
$4.56
$6.74
Upper estimate (@3%)
$7.04
$6.74
$6.34
$8.13
Low estimate
(@5%)
$3.72
$3.26
$2.67
$4.82
Best estimate
(@5%)
$4.74
$4.40
$3.89
$5.75
Upper estimate (@5%)
$6.01
$5.75
$5.42
$6.94
Low estimate
(@6%)
$3.46
$3.03
$2.48
$4.48
Best estimate
(@6%)
$4.40
$4.09
$3.62
$5.35
Upper estimate (@6%)
$5.58
$5.35
$5.03
$6.45
* Estimates derived using a full choice model not the simple multiplication of attribute values A
Discount rates shown in parentheses.
48
Chapter 7: Transferability of value estimates 7.1 Overview This Chapter of the report presents results from the questionnaires that asked city and regional households to make choices between alternative options for resource use in each case-study region. The results demonstrate that implicit price estimates for environmental and social attributes are significantly higher when attributes are presented to respondents for valuation in a regional context as opposed to a national context. Furthermore, statistical tests indicate that there are significant differences between the case-study regions in terms of the values estimated for some attributes. These differences indicate that framing and population effects are influential in determining values. The results imply that care must be taken in transferring value estimates from one context to another. In total, four benefit transfer tests (BT tests) were performed to gain an insight into how values change across different populations and frames of reference. This chapter provides a detailed description of each test, together with a summary of the main results. 7.2 Benefit transfer tests BT Test 1: Transferability of estimates from a national to regional context
This test examines whether the implicit prices estimated for attributes in the national context are equivalent to values estimated for the same set of attributes in a regional context. The test also establishes the magnitude of differences, and hence the size of scaling adjustment that is required if the national set of implicit prices is to be transferred to a regional setting. The null and alternative hypotheses under investigation are: H0: IPn (NF,NP) = IPn (RFx,RPx) H1: IPn (NF,NP) ≠ IPn (RFx,RPx) where; •
IPn is the implicit price for attribute n;
•
NF,NP denotes the choice model based on the national frame and national population sample; and
•
RFxRPx denotes the choice model based on the regional frame and regional population for case study x. The two regional frames and (populations) are Great Southern (Albany) and Fitzroy Basin (Rockhampton).
The implicit prices derived from each of the three models are plotted in Figure 7.1, together with a 95 per cent confidence interval for each estimate (denoted by the error bar). The confidence intervals were calculated using a technique developed by Krinsky and Robb (1986). Implicit price estimates are deemed to be equivalent if the confidence intervals for each estimate overlap. Using this test criteria, the null hypothesis is rejected for all attributes and it is concluded that implicit prices from the regional studies are significantly higher than those estimated for the national study (by a factor of 2 to 26 times, depending on the attribute in 49
question). A number of factors could be responsible for the different value estimates because the case studies differ from the national study in terms of: •
the respondent's frame of reference for valuing attributes;
•
the population sampled5; and
•
the scope of changes being presented to respondents for valuation.
The results support the prior of regular embedding; that is, consumers place a lower value on attributes when framed in a wide, national context versus a narrow, local context. A scope effect could also be responsible for the value differences given that larger changes were presented to respondents in the national study. However, this test does not allow firm conclusions to be drawn about the predominant cause of the differences. BT test 2, the next test to be reported, serves to disentangle framing effects from population differences so that the influence framing can be assessed in isolation. BT Test 2: The relative importance of framing
This test examines the equality of implicit price estimates derived from the national and regional versions of the questionnaire that were issued to separate samples of the same regional population. The objective of this test is to gauge the extent of the framing effect. This is made possible because the two samples for each case study test are drawn from the same population, so population effects are controlled for. The null and alternative hypotheses are: H0: IPn (NF,RPx) = IPn (RFx,RPx) H1: IPn (NF,RPx) ≠ IPn (RFx,RPx) where; •
IPn is the implicit price for attribute n;
•
NF,RPx denotes the choice model based on the national frame and regional population sample x, being respondents from either Albany or Rockhampton.
•
RFxRPx denotes the choice model based on the regional frame and regional population for case study x.
Upon comparing the IP’s from the national and regional frame, the null hypothesis is rejected for all attributes. It is concluded that respondents have significantly higher values when attributes are framed in a regional context (Figure 7.2). The scale of differences is similar to the findings from BT Test 1, which suggests that framing effects (due to scope or context differences) is the primary factor affecting the value estimates rather than population effects.
5
Whilst some socio-economic and attitudinal characteristics of the different populations are 'controlled for' in the modelling process, a wide range of other population characteristics remain unexplained and exogenous to the model.
50
Figure 7.1: Attribute implicit prices examined under BT Test 1.
Species NP,NF Aesthetics Waterways Community
RP,RF (Alb,GSR)
RP,RF (Roc,FRB)
-$4.00
-$3.00
-$2.00
-$1.00
$0.00
$1.00
$2.00
$3.00
$4.00
Implicit price
Note: Values for non-significant attributes are not plotted. Confidence intervals are shown by the error bars.
Figure 7.2: Attribute implicit prices examined under BT Test 2
Species
RP,NF (Alb,NAT)
Aesthetics Waterways
RP,NF (Roc,NAT)
Community RP,RF (Alb,GSR)
RP,RF (Roc,FRB)
-$4.00
-$3.00
-$2.00
-$1.00
$0.00
$1.00
$2.00
$3.00
$4.00
Implicit price
Note: Values for non-significant attributes are not plotted. Confidence intervals are shown by the error bars.
51
BT Test 3: Consistency of values across case study regions
The objective of BT Test 3 is to determine whether attribute value estimates vary across case study regions. Whilst the same set of attributes are being evaluated in each case study, the frame in which these attributes are ‘embedded’ is substantially different. Furthermore, the characteristics of each case study population are likely to be different. Some of this variation in population characteristics is controlled for by the socioeconomic variables included in the utility functions but attitudinal differences remain unaccounted for. The test was performed for respondents from both city and regional populations. For example, the choice models estimated for the Fitzroy Basin using preference data from Rockhampton and Brisbane respondents were compared to the models estimated for the Great Southern using data from Albany and Perth respondents. The null and alternative hypotheses for each type of comparison are as follows: Regional respondents H0: IPn (RFA,RPA) = IPn (RFB,RPB) H1: IPn (RFA,RPA) ≠ IPn (RFB,RPB) Capital city respondents H0: IPn (RFA,CPA) = IPn (RFB,CPB) H1: IPn (RFA,CPA) ≠ IPn (RFB,CPB) where; •
IPn is the implicit price for attribute n;
•
RFA and RFB denote the regional frames for case studies A and B (being the Great Southern and Fitzroy Basin;
•
RPA and RPB denote the regional populations for case studies A and B;
•
CPA and CPB denotes the capital city populations for case studies A and B.
The results indicate that value estimates for some attributes in the Fitzroy Basin and the Great Southern are significantly different. For example, respondent households from Rockhampton hold significantly higher values for social impacts in their local region relative to the values held by Perth and Albany respondents for social impacts in the Great Southern (Figure 7.3). Conversely, species protection is not valued in the Fitzroy region but it is a significant attribute in the Great Southern. These disparities demonstrate that the value estimates obtained in one region do not necessarily reflect community values in a different region, although there is a degree of consistency for some attributes. BT Test 4: Consistency of values across city and regional respondents
The purpose of this test is to examine whether respondents living within a given case study region have different attribute values to people living outside the region in an adjacent capital city. Therefore, in this test the frame is fixed but the population is allowed to vary. The null 52
and alternative hypotheses are: H0: IPn (RFx,RPx) = IPn (RFx,CPx) H1: IPn (RFx,RPx) ≠ IPn (RFx,CPx) •
IPn is the implicit price for attribute n;
•
RFx,RPx denotes the choice model based on the regional frame and regional population corresponding to case study x; and
•
RFxCPx denotes the choice model based on the regional frame and capital city population corresponding to case study x.
The results indicate that, with the exception of the social attribute, implicit prices for the attributes are statistically equivalent for regional and city households (Figure 7.3). In the case of social impacts, regional households in the Fitzroy Basin study (ie Rockhampton) do have significantly higher values than households residing in Brisbane city. For the other attributes, the results imply that it is safe to aggregate IP estimates from respondents in regional areas to city populations within the same state. Importantly, there is no evidence of values declining with distance from either of the case study regions. Parochialism does not appear to have played a significant role in influencing values in the regional communities. Figure 7.3: Attribute implicit prices examined under BT Tests 3 and 4.
Species
RP,RF (Alb,GSR)
Aesthetics Waterways
CP,RF (Per,GSR)
Community RP,RF (Roc,FRB)
CP,RF (Bris,FRB)
-$4.00
-$3.00
-$2.00
-$1.00
$0.00
$1.00
$2.00
$3.00
$4.00
Implicit price
Note: Values for non-significant attributes are not plotted. Confidence intervals are shown by the error bars.
53
7.3 Conclusions The most notable result obtained from the benefit transfer testing is the impact that framing has on attribute values. The results show unequivocally that implicit price estimates sourced from the national study are lower than those derived from the regional case studies. One possible reason for the value differences is embedding. That is, respondents could be cognisant of a larger array of environmental issues in the national frame and, hence, associate smaller values to the attributes under investigation. Alternatively, a scope effect could be responsible, meaning that the small changes in attribute levels presented to respondents in the case study questionnaires are valued more highly at the margin than the large changes in the national study. Regardless of which factor is the dominant reason for the value differences, household value estimates from the national study should be scaled up if they are to be validly transferred to a regional policy context. Guidelines for undertaking this transfer are contained in Chapter 8. Other key results from the case study analysis include: •
for both case studies, the geographic extent of the market for environmental values appears to extend beyond regional areas to include city populations. This finding holds for resource use changes in the local (state) context, but does not hold for changes in the national context where significant differences in values were estimated for city and rural populations;
•
the values estimated for social impacts appear to be less amenable for transfer, at least in Queensland where regional respondents (from Rockhampton) have significantly higher values for social impacts than city respondents (from Brisbane);
•
attribute values held by people in one region do not necessarily reflect community values in a different region (for the same set of attributes), although there is a degree of consistency for Landscape Aesthetics and Waterway Health.
54
Chapter 8: Benefit transfer guidelines 8.1 Overview The attribute implicit prices estimated in this non-market valuation study are useful for making a 'first pass' assessment of the size of non-market values associated with policies that have particular environmental and social impacts. The estimates are suitable for establishing the impacts of management decisions that affect major regions or the nation as a whole, and that can be described using one or more of the generic attributes. That is, the estimates can be used wherever impacts can be described in terms of changes in; •
the number of species protected;
•
the hectares of farmland repaired or bush protected;
•
the kilometres of river restored for recreation; and
•
the size of rural population.
The estimates are inappropriate for assessing impacts at the individual catchment level, or for valuing resource use changes that have very narrow and specific outcomes. Nor are the estimates suitable for determining the impact of policies that affect environmental assets that are considered to be national or regional 'icons', such as the protection of Koalas. The guidelines in Section 8.2 demonstrate how the implicit price estimates can be used to evaluate the non-market impacts of different policies. In circumstances where a more detailed and accurate assessment is warranted, the choice models estimated for the national study and regional case-study regions can be used to evaluate the welfare impacts (compensating surplus) of alternative scenarios. This more comprehensive approach was used to evaluate the resource use scenarios in Chapter 7. Guidelines for applying this more comprehensive approach to estimating welfare impacts is given Section 8.3. 8.2 Implicit price transfer Step 1: Defining the policy context
The first step is to determine whether the management policy is targeted at a particular region or whether it involves projects Australia-wide. If resource-use policies involve changes at a national level, then the set of attribute values estimated using the national sample of households is appropriate. For policies that are targeted at either of the two case study regions, it is recommended that the implicit prices estimated for these regions be used (see Appendix B for a complete tabulation of IP estimates). For regional assessments that do not correspond to one of the case study regions, it will be necessary to use the national estimates and calibrate the IP’s so that the values are appropriate for the region under investigation. A set of scaling factors for performing this calibration is given in Table 8.1. A range of scaling factors is given for each attribute to allow for a margin of variability between different regions and populations.
55
Table 8.1: Scaling factors for calibrating national value estimates to a regional context Attribute
National Implicit prices ($)
Scaling Factors
Species Protection
0.68
x2
Landscape Aesthetics
0.07
x 20-25
Waterway Health
0.08
x 20-25
Social impact
-0.09
x 6-26
Step 2: Defining the attribute changes
This step involves determining which attributes are impacted by the policy under investigation, and identifying the expected change in the attribute levels over a given time period relative to a 'business as usual' policy. Step 3: Aggregating the attribute values
Each attribute change caused by a particular policy (defined in Step 2) is then multiplied by its scaled implicit price (defined in Step 1). These so-calculated attribute values are then summed to yield an approximation of the average annual per household benefit to be derived from the implementation of the proposed policy. Step 4: Defining the target population
If the policy under investigation involves resource use changes at a national level, then the appropriate population for aggregating implicit prices is the population of Australian households. The impacts of changes implemented in particular regions should be restricted to the rural and city populations adjacent to the region in question. Extrapolation of values to other populations is speculative and not recommended. Step 5: Aggregation
It is recommended that the annual household values be aggregated to 45 per cent of the target population. If the analysis calls for an estimate of the full impact of a resource use change over a number of years, the annual values will need to be consolidated to a lump sum present value. A discount rate of 3 to 5 per cent is recommended. A regional policy assessment example
Consider the case of a proposal to redress land and water degradation in a region located in NSW. Under the proposal, 20,000 hectares of rural land will be rehabilitated, and 160 km of waterways will be restored. Analysis of the policy proposal by scientists indicates that the policy will ensure that three (3) additional species will be protected. Furthermore, it is predicted that 50 additional people per annum will leave the region because of the lower farming intensities the proposal involves. As a regional project, the implicit prices to be used in the valuation exercise will be scaled from the national estimates. Using the lower bound scaling factors in Table 8.1, the best estimate implicit prices are: •
Species Protection=
0.68 * 2 = $1.36 per species; 56
•
Landscape Aesthetics
0.07 * 20 = $1.40 per ten thousand hectares;
•
Waterway Health =
0.08 * 20 = $1.60 per 10 kilometres;
•
Social Impact
−0.09 * 6 = − $0.54 per 10 persons leaving each year
=
Given the changes in attribute levels specified, the best estimate of the community’s annual willingness to pay for the scenario is: (1.36 * 3) + (1.40 * 2) + (1.60 * 16) + (−0.54 * 5) = $29.78 per household This estimate is the amount, on average, that a household is willing to pay each year for twenty years to see the project proposed implemented. To estimate an aggregate value it is necessary to multiply the household value by an estimate of the size of the relevant population. This process includes making an adjustment to the survey estimates, via an aggregation factor, to allow for non-respondents in the sample. The following assumptions are used in this example: •
the relevant population includes metropolitan Sydney and proximate areas of rural NSW, which amounts to four million persons;
•
the number of people per household is 2.5;
•
the aggregation factor is 45 per cent.
Based on these assumptions, the best estimate of annual value would be: $29.78 * (4,000,000/2.5) * 0.45 = $21,441,600 per annum for 20 years. Where it becomes clear that the magnitude of the value estimated using this procedure is critical in the assessment of a policy, a more detailed analysis may be required. That analysis in the first instance may involve a refinement of the scaling factors used. By gaining a better understanding of the characteristics of the population to be affected by the policy under consideration, it can be assessed if the situation is closer to the Fitzroy Basin or the Great Southern case studies. Further analysis may also involve the use of a complete choice model rather than the aggregation of attribute values. As a general rule, if the project is justified when lower bound estimates are used, one can be very confident in recommending the project be accepted. Conversely, if a project can be justified only if the best estimate is used, then more analysis is probably needed. 8.3 Choice model transfer When the changes in attribute levels are relatively large, a more accurate estimate of changes in welfare can be obtained using the full choice model. This welfare measure is known as ‘compensating surplus’ and represents the total value of a change in the levels of multiple attributes away from the business as usual scenario. Use of the full choice model incorporates the impacts of the attributes, as well as the factors influencing choice that have not been defined in the choice sets. If a comprehensive assessment of welfare impacts is sought for changes in resource use at a regional level, it is recommended that one of the case study models should be employed for 57
benefit transfer. Tests show that both of the regional models - estimated with data from the corresponding regional population (ie Albany or Rockhampton) - produce the same welfare estimates for a standard change scenario. However, the Great Southern model yields estimates with a smaller error variability. Furthermore, all attributes in this model are statistically significant, while the insignificance of Species in the Fitzroy model is problematic. For these reasons, the Great Southern model is the preferred model for benefit transfer. The following checklist provides a guide to the procedure that should be followed when transferring the Great Southern model to a different region: q
Determine whether the set of attributes employed in this study adequately describe the issues in the target region and the policy outcomes that are under investigation.
q
Ensure that the ranges for the attribute levels in the target region are within the ranges used in the Great Southern questionnaire. Extrapolation outside these ranges will introduce transfer error.
q
Specify levels for the attributes that are appropriate for the region and the scenarios of interest. A business as usual scenario should be established as a benchmark against which to compare alternative management strategies.
q
Identify the target population for transfer. Ensure that the target population has attitudes and characteristics that are fundamentally similar to those used in the case study. It is recommended that the target population reside within the same state as the region under investigation. That is, the Great Southern model can be transferred to regions in other states, but the value estimates should only be aggregated to that state’s own population. Extrapolation of benefits to other States is speculative. An exception may be the situation where the target region straddles the border of two adjoining states.
q
Determine the mean socioeconomic characteristics of the target population. Two important characteristics include household annual income (before tax) and age. Substitute these mean values into the Great Southern model. The estimated parameters for this model are provided in Table 5.11.
q
Refer to Chapter 6 for technical details on how to calculate estimates of welfare change for a specific scenario relative to the status quo (see Box 6.1). For the Great Southern model, the error variability associated with these estimates is plus 85% and minus 64% of the mean value.
q
Aggregate the resultant household welfare estimates to 45 per cent of the target household population. The target population should be restricted to the rural and city populations adjacent to the region in question. Extrapolation of values to other populations is speculative and not recommended.
q
If the analysis calls for an estimate of the full impact of a resource use change over a number of years, the annual values will need to be consolidated to a lump sum present value. A discount rate of 3 to 5 per cent is recommended.
58
References Australia State of the Environment (1996). State of the Environment Advisory Council, Canberra. Australian Bureau of Statistics (1998). Education and training. Catalogue 4224.0. Australian Bureau of Statistics (1998). Income distribution. Catalogue 6253.0. Australian Bureau of Statistics (1998). Population projections. Catalogue 3222.0. Australian Bureau of Statistics (1999). Environmental issues: People’s views and practices. Catalogue 4602.0 Australian Bureau of Statistics (1999). Population by age and sex, Australian States and Territories. Catalogue 3201.0. Australian Bureau of Statistics (2000). Australian demographic statistics, March quarter. Catalogue 3101.0. Australian Bureau of Statistics (2000). Migration, Australia. Catalogue 3412.0. Australian Bureau of Statistics (2000). Regional population growth, Australia. Catalogue 3218.0. Australian National Parks and Wildlife Service (1992). Australian national strategy for the conservation of Australian species and communities threatened with extinction. Commonwealth of Australia, Canberra. Bennett, J. W. (1999). Some fundamentals of environmental choice modelling. Choice Modelling Research Reports, Report No. 11, University of New South Wales, Canberra. Blamey, R. K., Bennett, J. W., Louviere, J. J., Morrison, M. D., and Rolfe, J. C. (1999). The use of policy labels in environmental choice modelling studies. Choice Modelling Research Reports, Report No. 9. University of New South Wales, Canberra. Blamey, R. K., Bennett, J. W., Morrison, M. D., Louviere, J. J. and Rolfe, J. C. (1998). Attribute selection in environmental choice modelling studies: The effect of causally prior attributes. Choice Modelling Research Reports, Report No. 7. University of New South Wales, Canberra. Brower, R. (2000). Environmental value transfer: State of the art and future prospects. Ecological Economics, 32, pp. 137-152 Centre for International Economics (1997). A study to assess environmental values associated with water supply options. A report commissioned by ACTEW Corporation, Canberra. Fitzroy Basin Association (1998). Central Queensland strategy for sustainability, Rockhampton. Greene, W (1995). LIMDEP Econometric Software Inc. 59
Hundloe, T. J., McDonald, G. T. and Blamey, R. K. (1990). Socioeconomic analysis of nonextractive resource use in the Great Sandy Region. Institute of Applied Environmental Research, Griffith University. Imber, D., Stevenson. G., and Wilks. L. (1991). A contingent valuation survey of the Kakadu Conservation Zone, Resource Assessment Commission Research Paper No. 3, Commonwealth Government Printer, Canberra. Krinsky, I and Robb, A. L., (1986), ‘On approximating the statistical properties of elasticities’. Review of Economics and Statistics, 72: pp. 189-190. McFadden, D. (1974). The measurement of urban travel demand. Journal of Public Economics, 3, pp 303-328. Morrison, M., Bennett,J. and Blamey, R. (1999). Valuing Improved water Quality Using Choice Modelling. Water Resources Research, 35(9), pp 2805-2814. Pate, J. and Loomis, J. (1997). The effect of distance on willingness to pay values: A case study of wetlands and salmon in California. Ecological Economics, Vol. 20, pp 199-207 Queensland State of the Environment Report (1999). Queensland Environmental Protection Agency, Brisbane. Rolfe, J. and Bennett, J. (2000). Testing for framing effects in environmental choice modelling. Choice Modelling Research Reports, Report No. 13, University of New South Wales, Canberra. Science, Engineering, and Innovation Council (1999). Moving forward in natural resource management, Canberra. Sinden, J. A. (1987). Community support for soil conservation. Search, 18(4): 188-194. Standing Committee of Agriculture and Resource Management (2000). Managing natural resources in rural Australia for a sustainable future, Canberra. Sutherland, R. J. and Walsh, R. (1985). Effect of distance on the preservation value of water quality. Land Economics, Vol. 61, pp. 281-291. Western Australian Salinity Strategy (2000). State Salinity Council, Perth, Western Australia.
60
Appendix A: Script for the focus group discussions
1. Framing of environmental issues in the wider context. •
What do you think some of the issues are that are being faced by Australian society? What issues are of most concern to you? Please take a moment to write them down on your pad.
•
Out of the list on the board, how would you rank the issues in terms of their importance? What ranking would you give to the environment?
2. Awareness of environmental issues •
What are the first things that come to mind when we talk about environmental problems in Australia? I will give you a moment to list your ideas down on your pad. Take your time.
•
How would you rank these concerns in order of their relative importance? That is, what are the most pressing environmental problems in Australia?
•
In what ways do you think that your concerns about environmental issues are influenced by what you see in your own local area and state, as opposed to other regions of Australia?
•
Over the last 10 years do you think the overall quality of the environment in Australia has declined, improved or stayed much the same?
3. Land and water degradation: Attribute definition. Tonight I want to focus specifically on issues relating to land and water quality in Australia. Obviously the health of land and water is important for food production and the supply of fresh drinking water. But I want you to think about the other ways in which the environment is important to you. •
What specific factors or aspects of the environment do you think are important?
•
What factors of the environment would you like to see kept protected/preserved for your children’s children?
•
Suppose the government was to make additional funds available for addressing environmental problems. What evidence would convince you that the money was being well targeted and successful at improving environmental quality?4.
4. Responsibility and funding mechanisms •
If Australia’s environmental problems are to be adequately addressed, more money will need to be raised. How do you think environmental programs should be funded?
•
In reality, how do you expect environmental programs will be funded into the future?
•
If you were asked to support a proposal to increase the amount of public money spent on the environment, what information would you like to know before you made your decision?
61
Appendix B: Attribute implicit prices
Implicit prices for attributes, estimated for different combinations of population and (frame) Frame
National
National
National
Population
National
Albany
Rockhampton Albany
Perth
Rockhampton Brisbane
SPECIES
$ per species protected
mean
$0.68
$0.28
$1.27
NS
NS
$0.27
Great Sthn
$1.55
Great Sthn Fitzroy
Fitzroy
plus
$0.20
$0.24
$0.30
$0.78
$0.58
minus
$0.21
$0.30
$0.25
$0.67
$0.57
LOOK
$ per 10,000 ha of land restored
mean
$0.07
$0.21
$0.20
$1.84
$1.40
$1.57
$1.30
plus
$0.07
$0.08
$0.10
$0.95
$0.70
$1.68
$0.98
minus
$0.05
$0.07
$0.08
$0.78
$0.60
$1.16
$0.76
WATER
$ per 10km of waterways restored
mean
$0.08
$0.07
$1.56
$0.91
$2.02
$0.79
plus
$0.05
$0.07
$0.84
$0.61
$1.53
$0.70
minus
$0.04
$0.06
$0.64
$0.49
$1.08
$0.58
SOCIAL
$ per 10 persons migrating from rural areas
mean
-$0.09
-$0.11
-$0.06
-$0.55
-$0.71
-$2.24
-$1.03
plus
$0.02
$0.03
$0.02
$0.25
$0.20
$0.69
$0.36
minus
$0.02
$0.03
$0.03
$0.33
$0.26
$1.08
$0.42
NS
Notes: NS denotes attribute not statistically significant in the choice model.
62
Appendix C: Background information accompanying the national survey
63
64
65
66
